arch/arm64/mm/mmu.c

   1 /*
   2  * Based on arch/arm/mm/mmu.c
   3  *
   4  * Copyright (C) 1995-2005 Russell King
   5  * Copyright (C) 2012 ARM Ltd.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include <linux/export.h>
  21 #include <linux/kernel.h>
  22 #include <linux/errno.h>
  23 #include <linux/init.h>
  24 #include <linux/mman.h>
  25 #include <linux/nodemask.h>
  26 #include <linux/memblock.h>
  27 #include <linux/fs.h>
  28 #include <linux/io.h>
  29
  30 #include <asm/cputype.h>
  31 #include <asm/sections.h>
  32 #include <asm/setup.h>
  33 #include <asm/sizes.h>
  34 #include <asm/tlb.h>
  35 #include <asm/mmu_context.h>
  36 #include <mach/mtk_memcfg.h>
  37
  38 #include "mm.h"
  39
  40 /*
  41  * Empty_zero_page is a special page that is used for zero-initialized data
  42  * and COW.
  43  */
  44 struct page *empty_zero_page;
  45 EXPORT_SYMBOL(empty_zero_page);
  46
  47 pgprot_t pgprot_default;
  48 EXPORT_SYMBOL(pgprot_default);
  49
  50 static pmdval_t prot_sect_kernel;
  51
  52 struct cachepolicy {
  53         const char      policy[16];
  54         u64             mair;
  55         u64             tcr;
  56 };
  57
  58 static struct cachepolicy cache_policies[] __initdata = {
  59         {
  60                 .policy         = "uncached",
  61                 .mair           = 0x44,                 /* inner, outer non-cacheable */
  62                 .tcr            = TCR_IRGN_NC | TCR_ORGN_NC,
  63         }, {
  64                 .policy         = "writethrough",
  65                 .mair           = 0xaa,                 /* inner, outer write-through, read-allocate */
  66                 .tcr            = TCR_IRGN_WT | TCR_ORGN_WT,
  67         }, {
  68                 .policy         = "writeback",
  69                 .mair           = 0xee,                 /* inner, outer write-back, read-allocate */
  70                 .tcr            = TCR_IRGN_WBnWA | TCR_ORGN_WBnWA,
  71         }
  72 };
  73
  74 /*
  75  * These are useful for identifying cache coherency problems by allowing the
  76  * cache or the cache and writebuffer to be turned off. It changes the Normal
  77  * memory caching attributes in the MAIR_EL1 register.
  78  */
  79 static int __init early_cachepolicy(char *p)
  80 {
  81         int i;
  82         u64 tmp;
  83
  84         for (i = 0; i < ARRAY_SIZE(cache_policies); i++) {
  85                 int len = strlen(cache_policies[i].policy);
  86
  87                 if (memcmp(p, cache_policies[i].policy, len) == 0)
  88                         break;
  89         }
  90         if (i == ARRAY_SIZE(cache_policies)) {
  91                 pr_err("ERROR: unknown or unsupported cache policy: %s\n", p);
  92                 return 0;
  93         }
  94
  95         flush_cache_all();
  96
  97         /*
  98          * Modify MT_NORMAL attributes in MAIR_EL1.
  99          */
 100         asm volatile(
 101         "       mrs     %0, mair_el1\n"
 102         "       bfi     %0, %1, #%2, #8\n"
 103         "       msr     mair_el1, %0\n"
 104         "       isb\n"
 105         : "=&r" (tmp)
 106         : "r" (cache_policies[i].mair), "i" (MT_NORMAL * 8));
 107
 108         /*
 109          * Modify TCR PTW cacheability attributes.
 110          */
 111         asm volatile(
 112         "       mrs     %0, tcr_el1\n"
 113         "       bic     %0, %0, %2\n"
 114         "       orr     %0, %0, %1\n"
 115         "       msr     tcr_el1, %0\n"
 116         "       isb\n"
 117         : "=&r" (tmp)
 118         : "r" (cache_policies[i].tcr), "r" (TCR_IRGN_MASK | TCR_ORGN_MASK));
 119
 120         flush_cache_all();
 121
 122         return 0;
 123 }
 124 early_param("cachepolicy", early_cachepolicy);
 125
 126 /*
 127  * Adjust the PMD section entries according to the CPU in use.
 128  */
 129 static void __init init_mem_pgprot(void)
 130 {
 131         pteval_t default_pgprot;
 132         int i;
 133
 134         default_pgprot = PTE_ATTRINDX(MT_NORMAL);
 135         prot_sect_kernel = PMD_TYPE_SECT | PMD_SECT_AF | PMD_ATTRINDX(MT_NORMAL);
 136
 137 #ifdef CONFIG_SMP
 138         /*
 139          * Mark memory with the "shared" attribute for SMP systems
 140          */
 141         default_pgprot |= PTE_SHARED;
 142         prot_sect_kernel |= PMD_SECT_S;
 143 #endif
 144
 145         for (i = 0; i < 16; i++) {
 146                 unsigned long v = pgprot_val(protection_map[i]);
 147                 protection_map[i] = __pgprot(v | default_pgprot);
 148         }
 149
 150         pgprot_default = __pgprot(PTE_TYPE_PAGE | PTE_AF | default_pgprot);
 151 }
 152
 153 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 154                               unsigned long size, pgprot_t vma_prot)
 155 {
 156         if (!pfn_valid(pfn))
 157                 return pgprot_noncached(vma_prot);
 158         else if (file->f_flags & O_SYNC)
 159                 return pgprot_writecombine(vma_prot);
 160         return vma_prot;
 161 }
 162 EXPORT_SYMBOL(phys_mem_access_prot);
 163
 164 static void __init *early_alloc(unsigned long sz)
 165 {
 166         void *ptr = __va(memblock_alloc(sz, sz));
 167         memset(ptr, 0, sz);
 168         return ptr;
 169 }
 170
 171 static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
 172                                   unsigned long end, unsigned long pfn)
 173 {
 174         pte_t *pte;
 175
 176         if (pmd_none(*pmd)) {
 177                 pte = early_alloc(PTRS_PER_PTE * sizeof(pte_t));
 178                 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
 179         }
 180         BUG_ON(pmd_bad(*pmd));
 181
 182         pte = pte_offset_kernel(pmd, addr);
 183         do {
 184                 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 185                 pfn++;
 186         } while (pte++, addr += PAGE_SIZE, addr != end);
 187 }
 188
 189 static void __init alloc_init_pmd(pud_t *pud, unsigned long addr,
 190                                   unsigned long end, phys_addr_t phys)
 191 {
 192         pmd_t *pmd;
 193         unsigned long next;
 194
 195         /*
 196          * Check for initial section mappings in the pgd/pud and remove them.
 197          */
 198         if (pud_none(*pud) || pud_bad(*pud)) {
 199                 pmd = early_alloc(PTRS_PER_PMD * sizeof(pmd_t));
 200                 pud_populate(&init_mm, pud, pmd);
 201         }
 202
 203         pmd = pmd_offset(pud, addr);
 204         do {
 205                 next = pmd_addr_end(addr, end);
 206                 /* try section mapping first */
 207                 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
 208                         pmd_t old_pmd =*pmd;
 209                         set_pmd(pmd, __pmd(phys | prot_sect_kernel));
 210                         /*
 211                          * Check for previous table entries created during
 212                          * boot (__create_page_tables) and flush them.
 213                          */
 214                         if (!pmd_none(old_pmd))
 215                                 flush_tlb_all();
 216                 } else {
 217                         alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys));
 218                 }
 219                 phys += next - addr;
 220         } while (pmd++, addr = next, addr != end);
 221 }
 222
 223 static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
 224                                   unsigned long end, unsigned long phys)
 225 {
 226         pud_t *pud = pud_offset(pgd, addr);
 227         unsigned long next;
 228
 229         do {
 230                 next = pud_addr_end(addr, end);
 231                 alloc_init_pmd(pud, addr, next, phys);
 232                 phys += next - addr;
 233         } while (pud++, addr = next, addr != end);
 234 }
 235
 236 /*
 237  * Create the page directory entries and any necessary page tables for the
 238  * mapping specified by 'md'.
 239  */
 240 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
 241                                   phys_addr_t size)
 242 {
 243         unsigned long addr, length, end, next;
 244         pgd_t *pgd;
 245
 246         if (virt < VMALLOC_START) {
 247                 pr_warning("BUG: not creating mapping for 0x%016llx at 0x%016lx - outside kernel range\n",
 248                            phys, virt);
 249                 return;
 250         }
 251
 252         addr = virt & PAGE_MASK;
 253         length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
 254
 255         pgd = pgd_offset_k(addr);
 256         end = addr + length;
 257         do {
 258                 next = pgd_addr_end(addr, end);
 259                 alloc_init_pud(pgd, addr, next, phys);
 260                 phys += next - addr;
 261         } while (pgd++, addr = next, addr != end);
 262 }
 263
 264 #ifdef CONFIG_EARLY_PRINTK
 265 /*
 266  * Create an early I/O mapping using the pgd/pmd entries already populated
 267  * in head.S as this function is called too early to allocated any memory. The
 268  * mapping size is 2MB with 4KB pages or 64KB or 64KB pages.
 269  */
 270 void __iomem * __init early_io_map(phys_addr_t phys, unsigned long virt)
 271 {
 272         unsigned long size, mask;
 273         bool page64k = IS_ENABLED(CONFIG_ARM64_64K_PAGES);
 274         pgd_t *pgd;
 275         pud_t *pud;
 276         pmd_t *pmd;
 277         pte_t *pte;
 278
 279         /*
 280          * No early pte entries with !ARM64_64K_PAGES configuration, so using
 281          * sections (pmd).
 282          */
 283         size = page64k ? PAGE_SIZE : SECTION_SIZE;
 284         mask = ~(size - 1);
 285
 286         pgd = pgd_offset_k(virt);
 287         pud = pud_offset(pgd, virt);
 288         if (pud_none(*pud))
 289                 return NULL;
 290         pmd = pmd_offset(pud, virt);
 291
 292         if (page64k) {
 293                 if (pmd_none(*pmd))
 294                         return NULL;
 295                 pte = pte_offset_kernel(pmd, virt);
 296                 set_pte(pte, __pte((phys & mask) | PROT_DEVICE_nGnRE));
 297         } else {
 298                 set_pmd(pmd, __pmd((phys & mask) | PROT_SECT_DEVICE_nGnRE));
 299         }
 300
 301         return (void __iomem *)((virt & mask) + (phys & ~mask));
 302 }
 303 #endif
 304
 305 static void __init map_mem(void)
 306 {
 307         struct memblock_region *reg;
 308         phys_addr_t limit;
 309
 310         /*
 311          * Temporarily limit the memblock range. We need to do this as
 312          * create_mapping requires puds, pmds and ptes to be allocated from
 313          * memory addressable from the initial direct kernel mapping.
 314          *
 315          * The initial direct kernel mapping, located at swapper_pg_dir, gives
 316          * us PUD_SIZE (4K pages) or PMD_SIZE (64K pages) memory starting from
 317          * PHYS_OFFSET (which must be aligned to 2MB as per
 318          * Documentation/arm64/booting.txt).
 319          */
 320         if (IS_ENABLED(CONFIG_ARM64_64K_PAGES))
 321                 limit = PHYS_OFFSET + PMD_SIZE;
 322         else
 323                 limit = PHYS_OFFSET + PUD_SIZE;
 324         memblock_set_current_limit(limit);
 325
 326         /* map all the memory banks */
 327         for_each_memblock(memory, reg) {
 328                 phys_addr_t start = reg->base;
 329                 phys_addr_t end = start + reg->size;
 330                 MTK_MEMCFG_LOG_AND_PRINTK(KERN_ALERT
 331                         "[PHY layout]kernel   :   "
 332                         "0x%08llx - 0x%08llx (0x%llx)\n",
 333                         (unsigned long long)start,
 334                         (unsigned long long)start + reg->size - 1,
 335                         (unsigned long long)reg->size);
 336
 337                 if (start >= end)
 338                         break;
 339
 340 #ifndef CONFIG_ARM64_64K_PAGES
 341                 /*
 342                  * For the first memory bank align the start address and
 343                  * current memblock limit to prevent create_mapping() from
 344                  * allocating pte page tables from unmapped memory.
 345                  * When 64K pages are enabled, the pte page table for the
 346                  * first PGDIR_SIZE is already present in swapper_pg_dir.
 347                  */
 348                 if (start < limit)
 349                         start = ALIGN(start, PMD_SIZE);
 350                 if (end < limit) {
 351                         limit = end & PMD_MASK;
 352                         memblock_set_current_limit(limit);
 353                 }
 354 #endif
 355
 356                 create_mapping(start, __phys_to_virt(start), end - start);
 357         }
 358
 359         /* Limit no longer required. */
 360         memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
 361 }
 362
 363 /*
 364  * paging_init() sets up the page tables, initialises the zone memory
 365  * maps and sets up the zero page.
 366  */
 367 void __init paging_init(void)
 368 {
 369         void *zero_page;
 370
 371         init_mem_pgprot();
 372         map_mem();
 373
 374         /*
 375          * Finally flush the caches and tlb to ensure that we're in a
 376          * consistent state.
 377          */
 378         flush_cache_all();
 379         flush_tlb_all();
 380
 381         /* allocate the zero page. */
 382         zero_page = early_alloc(PAGE_SIZE);
 383
 384         bootmem_init();
 385
 386         empty_zero_page = virt_to_page(zero_page);
 387
 388         /* Ensure the zero page is visible to the page table walker */
 389         dsb();
 390
 391         /*
 392          * TTBR0 is only used for the identity mapping at this stage. Make it
 393          * point to zero page to avoid speculatively fetching new entries.
 394          */
 395         cpu_set_reserved_ttbr0();
 396         flush_tlb_all();
 397 }
 398
 399 /*
 400  * Enable the identity mapping to allow the MMU disabling.
 401  */
 402 void setup_mm_for_reboot(void)
 403 {
 404         cpu_switch_mm(idmap_pg_dir, &init_mm);
 405         flush_tlb_all();
 406 }
 407
 408 /*
 409  * Check whether a kernel address is valid (derived from arch/x86/).
 410  */
 411 int kern_addr_valid(unsigned long addr)
 412 {
 413         pgd_t *pgd;
 414         pud_t *pud;
 415         pmd_t *pmd;
 416         pte_t *pte;
 417
 418         if ((((long)addr) >> VA_BITS) != -1UL)
 419                 return 0;
 420
 421         pgd = pgd_offset_k(addr);
 422         if (pgd_none(*pgd))
 423                 return 0;
 424
 425         pud = pud_offset(pgd, addr);
 426         if (pud_none(*pud))
 427                 return 0;
 428
 429         pmd = pmd_offset(pud, addr);
 430         if (pmd_none(*pmd))
 431                 return 0;
 432
 433         pte = pte_offset_kernel(pmd, addr);
 434         if (pte_none(*pte))
 435                 return 0;
 436
 437         return pfn_valid(pte_pfn(*pte));
 438 }
 439 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 440 #ifdef CONFIG_ARM64_64K_PAGES
 441 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 442 {
 443         return vmemmap_populate_basepages(start, end, node);
 444 }
 445 #else   /* !CONFIG_ARM64_64K_PAGES */
 446 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 447 {
 448         unsigned long addr = start;
 449         unsigned long next;
 450         pgd_t *pgd;
 451         pud_t *pud;
 452         pmd_t *pmd;
 453
 454         do {
 455                 next = pmd_addr_end(addr, end);
 456
 457                 pgd = vmemmap_pgd_populate(addr, node);
 458                 if (!pgd)
 459                         return -ENOMEM;
 460
 461                 pud = vmemmap_pud_populate(pgd, addr, node);
 462                 if (!pud)
 463                         return -ENOMEM;
 464
 465                 pmd = pmd_offset(pud, addr);
 466                 if (pmd_none(*pmd)) {
 467                         void *p = NULL;
 468
 469                         p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 470                         if (!p)
 471                                 return -ENOMEM;
 472
 473                         set_pmd(pmd, __pmd(__pa(p) | prot_sect_kernel));
 474                 } else
 475                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 476         } while (addr = next, addr != end);
 477
 478         return 0;
 479 }
 480 #endif  /* CONFIG_ARM64_64K_PAGES */
 481 void vmemmap_free(unsigned long start, unsigned long end)
 482 {
 483 }
 484 #endif  /* CONFIG_SPARSEMEM_VMEMMAP */